Deep learning-guided production quality estimation for virtual environment-based applications

Abstract

In modern smart factories, quality estimation is vital for maximum productivity. However, quality estimation by definition relies on an imbalanced dataset, as most smart factories are highly efficient. In this research, we propose a guided quality estimation system that can recognize faulty data among a highly imbalanced production dataset. We also propose a customized LSTM model that is trained to ensure high accuracy in the quality estimation system. This is achieved by our proposed batch-wise balanced training method. Moreover, traditional means of evaluation for this type of method are not suitable, again due to the highly imbalanced nature of the dataset. Thus, a proper evaluation metric is also discussed. The proposed customized LSTM model with custom batch-wise SMOTE + ENN achieved 99.9% accuracy with an f(1) score of 95%. This new proposed method for the imbalanced smart factory quality estimation will improve drastically and give pathway to more improved quality. Finally, we discuss practical implementation for the edge server consisting of the proposed guided production estimation system and real-time visualization. Feasibility analysis of this virtual environment-based application of the proposed framework ensured low computational overhead and faster processing.